Fusion Diaries: Snowflake UDF fetching, access enforcement, upcoming arguments key for generic tests (2025-07-22)

[joellabes]

Hot off the presses, it's your weekly Fusion update.

• 66 issues closed across the public and internal repos
• 5 new betas released
• No more unnecessary authentication popup prompts for Danielle or Brian

Check below for bugfixes, upcoming changes to unsafe static analysis and generic test definitions, and a meme that proves that comedy is just tragedy bugs plus time better file lock management.

🆕 What’s been released this week

dbt Fusion engine releases

Run dbt system update to get the latest and greatest, including:

2.0.0-beta.40: detect unit test fixtures in subdirectories, fix issue where sources would be incorrectly marked as state:modified

2.0.0-beta.41: support updates_on: any in state-aware orchestration, allow env_var() to accept named arguments in profiles.yml

• Match Python behaviour for indent Jinja argument names and modules.datetime.timedelta() positional argument order
• Case-insensitive bool handling in Jinja
• Better BigQuery error messages for failed auth

2.0.0-beta.42: support some --no-prefixed command line flags, support compiling Snowflake array_agg with windowing and ordering

2.0.0-beta.43: Support overriding DBT_PACKAGE_HUB_URL env var, fixed issue where unit tests could block a run from completing

2.0.0-beta.44: fetch existing Snowflake UDFs so they can be used in static analysis (set DBT_ENGINE_EXPERIMENTAL_LIST_UDFS=1), warn when a view used as a source doesn’t have loaded_at_field or loaded_at_query (it will always be considered fresh)

• Support ceil and floor methods for the round Jinja filter
• Include accurate thread_ids in run_results.json
• Fix Jinja discrepancies when a var is undefined or a method is called inside of an expression instead of a statement
• Moved test utilities to the source-available repo, to simplify collaboration with partners
• Support compiling a Snowflake variant type to a timestamp_ntz
• Apply configs like enabled to sources when defined in dbt_project.yml

2.0.0-beta.45: enforce access model governance rules, prevent excessive Snowflake externalauth browser popups

• Make it possible to use the as_text filter on a Relation
• Performance improvements from reusing Jinja contexts
• Preparation for upcoming Redshift adapter release
• Fix BigQuery incremental models using partition_by config, and fix a related undeclared variable error
• Support {% for x in my_list[::-1] -%} syntax (but come on, just use {% for x in my_list | reverse %} please!)
• Don’t try to describe Databricks system tables with JSON output
• Fix panic when trying to statically analyse BQ external datasets

VS Code extension + LSP

• Fixed issue where LSP might not start due to port number conflict
• Better error message when access token verification fails
• Renaming a model will refactor any macros that contain the model
• Windows fixes: saving a file correctly triggers a recompile, improved LSP installation reliability

🚧 Coming up soon

Caching introspective query results for faster rendering

Projects with many introspective queries (including our internal analytics project!) see unpleasantly long compile times, due to the overhead of repeatedly fetching the results. We’re gonna cache ‘em!

Introspection will no longer disable static analysis

In an earlier Fusion Diary, we linked out to this Discourse discussion about static analysis and introspective queries. Based on this and feedback we’ve heard on calls with beta users, we’ve agreed that the old system was too strict and we’re changing it.

Previously, having an introspective query in your project would implicitly set static_analysis: off for that model and its children, and we required you to opt back into static analysis by setting static_analysis: unsafe. The dbt Fusion engine will soon step down to unsafe static analysis automatically as needed.

Check out a preview of the docs for more details.

Generic tests’ arguments are becoming strongly-typed

This is a big one - today, custom generic tests can be free-form YAML, which makes it impossible to statically validate configs and keyword arguments. This is the same sort of problem as plusless configs in dbt_project.yml or unknown config names in whatever.yml. In fact, you can think of this change as in the same way as moving user-defined configs to the meta dictionary.

By removing this ambiguity, we’ll be able to consistently render Jinja at the right time, solving issues like dbt-fusion#191 and dbt-fusion#250.

Here’s a before and after (expand to see code)

#before 
#models/my_models.yml 
models:
  - name: model_name
    columns:
      - name: num_web_sessions
        tests:
          - dbt_utils.accepted_range:
              min_value: 0
              inclusive: false
              config:
                where: "num_orders > 0"
                severity: warn

#after
#models/my_models.yml 
models:
  - name: model_name
    columns:
      - name: num_web_sessions
        tests:
          - dbt_utils.accepted_range:
              arguments:
                min_value: 0
                inclusive: false
              config:
                where: "num_orders > 0"
                severity: warn

Of course, dbt-autofix has your back on these, and we’re updating dbt Core to support the same syntax. None of this is live today, just a heads up.

🤔 Looking for feedback

@graciegoheen just dropped the GitHub discussion for native UDF definitions in the dbt authoring layer. This is different to the discovery of existing UDFs described above - this will enable dbt Core and the dbt Fusion engine to directly manage your UDFs.

There's a lot of open questions over there, so make sure you give it a read and share your thoughts.

🐉 Here be dragons

Occasionally, the VS Code extension get stuck in a bad state when it’s auto-updated. For now, restarting your IDE should wake it back up. Tracking in #420

🏁 Made it to the meme

We’re pretty confident that we’ve got them all now! But seriously: if you keep seeing multiple auth prompts on Snowflake, please let us know.

[joellabes]

Adding some extra context from @venkaa28 now that the arguments change is merged:

This PR removes the use of verbatim wrapping around vec for all properties, allowing all fields to be fully rendered when rendering node properties.

What this means: All Jinja used in data tests must now be resolvable at the time we evaluate the node—except for Jinja that is passed in as arguments to custom generic tests. These are still wrapped in verbatim, allowing late rendering.

Additionally, this PR consolidates all custom test arguments into a dedicated args field. This removes ambiguity between whether a parameter is a kwarg or a config value, which in turn enables strong typing for custom generic test structs.

This is behind a behaviour change flag in dbt Core:

If you don't want to see these warnings please put all configs under config and all custom/alternate format generic test arguments under arguments.

If you want autofix to pick up this behavior change, please add this to your dbt_project.yml

flags:
  require_generic_test_arguments_property: true

and run

uvx --from git+https://github.com/dbt-labs/dbt-autofix.git dbt-autofix deprecations